University AI safety groups have historically underinvested in policy talent pipelines compared to technical ones. This represents a significant gap in the talent funnel, primarily driven by Founder Effects (most groups were started by CS students optimizing for technical recruitment). We believe this is a high-leverage area for improvement, and at the margin, we prefer investing in strong policy programs over technical ones, particularly at universities with concentrations of top political or public policy talent.

The Case for Policy Fieldbuilding

Technical AI safety groups have demonstrated strong track records in placing students into full-time technical roles at organizations like METR, Anthropic, Apollo, Google DeepMind, Redwood Research, etc. However, the policy pipeline remains underdeveloped. This is beginning to change as more groups take policy fieldbuilding seriously, but significant room for impact remains.

The AI policy ecosystem needs capable people who deeply internalize catastrophic risk concerns and can operate effectively in policy contexts. Unlike technical roles, where institutional culture often ensures alignment by default, policy roles require explicit selection for people with strong value alignment and commitment to AI safety to be impactful.

Strategic Considerations

Policy outcomes are not net positive by default.

In the technical space, the opportunity landscape relatively quality-filtered. Organizations like Anthropic and AI safety nonprofits maintain strong safety cultures. Even moderately bought-in hires will work on impactful projects surrounded by people taking x-risk seriously.

The policy landscape operates differently:

• Larger opportunity surface area: Most high-impact policy positions don't explicitly involve AI safety.
• Higher variance in outcomes: Placing someone at a think tank, campaign, or government office only creates counterfactual value if they actively prioritize AI safety concerns.
• Potential for negative impact: Insufficiently aligned individuals in policy roles could actively increase AI risk through poor advocacy or by supporting legislation that doesn’t reduce AI risks (prioritizing innovation instead, for example).

Politics vs. Policy

Policy (research-focused):

• Think tanks, academic institutions, international bodies, lab policy teams
• Emphasis on analysis, position papers, institutional design
• Historically prioritized by the AI safety community due to research aptitude overlap

Politics (implementation-focused):

• Working with elected officials, staffers, campaigns, government agencies
• Emphasis on stakeholder relationships, political feasibility, coalition-building
• Neglected but increasingly valuable for shaping actual governance outcomes